What's new in 1.52.0 Send your next question while AI is still replying. Queue up to ten follow-ups in a saved chat, then edit or reorder them before they start. They run one at a time, even when you switch chats. • More room for your conversation on iPhone and iPad, with compact message headers, clearer menus, and simpler Send and Stop controls. • Pick up a paused answer where you left off. Continued text stays together as one answer, including when you copy or share it. • Fewer repeated tool searches and more reliable answers after tool use with supported local models. We've also made typing in App Help more responsive and fixed Agent Workflows settings so changes take effect immediately.

What's new in 1.51.0 • Keep track of replies across chats with activity shown in history and Continue for saved partial answers. Starting another local reply pauses the earlier one and saves its progress; cloud chats can run alongside each other within your device's limits. Regenerating an answer keeps the original until its replacement is saved successfully. Stop controls now stay available when switching cloud chats, and Thinking is kept separate from the answer. • Models are easier to remove from the model picker. You can also follow download progress and cancel a download, with speed and estimated time remaining shown when enough data is available. Download status is more reliable after reopening the app. • The model picker has fewer repeated status labels. On iPhone, the Last Used card stays out of the way while you browse and returns when you scroll back to the top, without flickering in and out. • Hugging Face GGUF file links now point to the download. Repository links direct you to the model browser, and small model sizes display correctly. The model library's navigation title is visible again on iPhone. • Free up storage after removing models without restarting the app. Model Storage updates which shared files are still needed and explains which installed models depend on them. • Chat input on iPhone and iPad keeps your draft and attachments when you expand or close it. Keyboard focus and transitions are smoother, including when you tap quickly or open the side menu. These changes help you return to unfinished chats and free up storage with fewer steps, so you can spend more time on the conversation.

Version 1.50.0 makes AI Chat more dependable while you move between conversations, and gives you more ways to transcribe on iPhone and iPad. • Start a reply, open another conversation, and return to find the response where it belongs. If a response is interrupted, you can retry from that conversation, with any saved partial output retained. • Local chat models now show which engine they use in the model picker, making similar versions easier to tell apart. We also updated the engine behind local chat for faster responses. • More transcription models are available on supported iPhone and iPad devices, including additional Parakeet, Canary, Nemotron, Cohere, and MOSS choices. Availability still depends on the device and model requirements. • Qwen3-ASR now offers four size and precision choices on eligible devices. You can remove an older download separately and choose a current one. Your recordings and transcripts stay where they are. • Voice Note on iPhone puts Record within easy reach and groups optional recording settings. Model details and download activity are clearer. Speech downloads show a Cancel button, and finished models are easier to find before you choose to use them. • Long transcription and speech jobs now use smaller chunks on supported local models to help keep memory use under control. The app still checks each model's limits. • Fixed Parakeet TDT v3 being marked incomplete after a successful download. On Mac, Canary 1B v2 now prepares without blocking the app and shows progress during audio imports.

Version 1.49.0 is for the work that starts with a long recording, a scanned document, or an idea you want to hear out loud. It also makes a growing app much easier to find your way around. • Recording a meeting, class, interview, or research session? MOSS Transcribe + Diarize is our recommended model for meeting notes. It separates speakers and now uses less memory, so long recordings can finish more reliably. Your original audio stays untouched. • Add a scanned report, paper, receipt, contract, or class handout to AI Chat and ask questions about it. If a PDF has no readable text, the app asks before running OCR on your device. You can follow progress page by page or cancel without leaving a half-finished attachment. • Supertonic, VoxCPM2 Int8, VoxCPM2 BF16, and IndexTTS2 are new in this release. Depending on the model and your device, you can make narration for a presentation, create a voiceover, listen back to your own writing, or use a saved reference voice. • The Menu now keeps the workspaces you use most under Pinned, with everything else under Available. Settings are grouped by purpose and adapt to the screen, whether you are on an iPhone or working with a sidebar on a larger device. • Agent Workflows are easier to set up. Start with an Agent Profile for a repeatable job, such as research or editing, then add the AI Skills, MCP servers, and tools it needs. On Mac, messaging app connections now have their own Settings area. • The Model Library is now the main place to browse, download, choose, repair, or remove chat, transcription, speech, and embedding models. Speech downloads also show clearer progress, including quiet stretches during large transfers. • Saved chats open correctly on the first try. Long answers scroll more smoothly, and you can begin reading before rich formatting is done. • Cloud requests have more time to respond. If one fails, the app stops cleanly, keeps any partial reply, and leaves a successfully loaded model ready for another try. • We also cleaned up the AI Chat model header and Voice Note controls. Conversation Mode returns to listening after it speaks, and deleting an active chat no longer lets old work bring it back.

Version 1.48.0 adds AI Browser, chats that survive interruptions, MCP servers, and much more control over your models and memory. • AI Browser: Browse with tabs, profiles, history, downloads, and restored sessions, then ask your current model to read the page, help you with it, or act on it. The app tells you what a site can see and asks before anything changes outside the app. • Browser tasks that finish: Switch models before a task, follow a clearer timeline of what the agent did, and resize the task pane on Mac. Page reading and action checks improved, the app notices when a task is going in circles, and results are better on long or complex sites. • Chats that survive interruptions: A reply keeps going and reconnects if you navigate away. Queue your next prompt with Send Next or Stop & Send, and compare several models against the same frozen input. • Agent Profiles, schedules, and MCP: Create agent workspaces that stick around, run them when you want, or put supported work on a schedule. Connect remote MCP tool servers and install reviewed AI Skills that cannot run code, while you stay in charge of which tools are on. • Settings for each agent: Give individual Flow Agent steps and Subagents their own temperature and context window, or let them inherit your conversation defaults. • One place for model storage: See how much space chat, embedding, speech, and voice models use. Recover interrupted downloads and safely review duplicate, legacy, incomplete, or orphaned files. • Global Memory you can manage: Search, filter, pin, edit, delete, and undo memories in their own settings area. When the AI wants to change a memory, it now asks the same way tools do. • Faster recovery: Reload your last chat model in one step, and keep a conversation ready when reading aloud borrows a speech engine or restores a model afterwards. • Help, settings, and tools: App Help understands plain questions and comparisons better, support drafts open in more email apps, Settings sections respond faster, and local GGUF and MLX models recognize tool calls more reliably.

Version 1.47.0 adds a new Listen Library, RSS Feeds, App Help, and widgets. • Listen Library: Turn PDFs, Markdown, text files, readable web articles, and scanned books into a personal reading and listening queue. Keep your place, manage Up Next, and use the new sleep timer when listening at night. • Scanned books and image text: Scan or import photographed pages, review and correct the on-device OCR, then save the result as a book that keeps each page beside its spoken text. Text to Speech can also capture a passage from an image. • RSS Feeds: Follow RSS, Atom, and JSON feeds inside the app, browse curated suggestions, and build a listening queue from sources you choose. • App Help: Ask how a feature or setting works in a focused help chat backed by the app's built-in guide. App Help can point you to the right settings while keeping help questions separate from your saved chats. • New widgets: Add Listen, Voice Note, Text Note, or Quick Actions widgets on iPhone, iPad, and Mac. Resume playback, play the next item, open RSS Feeds, or start a note with fewer taps. • Better multilingual support: Expanded interface translations and more consistent language handling throughout the app. Widgets follow the language selected in On-Device AI. • More model choices: The built-in catalog now includes more ready-to-download models, including Gemma 4 QAT options, Devstral Small 2, Nemotron 3 Nano, Qwythos. • Smoother speech workflows: Long Kokoro Speak Text shortcuts can begin playback sooner. Shortcuts also recognize your paid model access more reliably when they start the app in the background. • Fixes and polish: Improved Listen imports, page editing, OCR review, reader layouts, playback handoff, feed discovery, and App Help navigation.

Version 1.46.0 adds Calendar and Reminders tools, a faster Text Note workspace, stronger AI Search, and more Shortcuts support. • Calendar and Reminders tools: With your permission, AI tools can read, create, update, complete, move, and delete events or reminders after you review changes. • Text Note workspace: Create a note quickly, pick the Knowledge Library it belongs to, edit in a cleaner notes view, and keep it searchable. • Cleaner AI Search context: Web results now skip consent pages, cookie notices, search-engine boilerplate, and other unusable pages, with fallback behavior when a result set is noisy. • Save AI Search results: Import a completed AI Search answer into a Knowledge Library as a curated Markdown snapshot with the query, answer, sources, and supporting context. • Smarter note destinations: When you use multiple Knowledge Libraries, the app asks where a new note should go instead of creating duplicate default libraries. • Customizable side menu: Choose which sections stay pinned in the sidebar and open everything else from the new Menu. • More Shortcuts actions: Transcribe audio files, speak text, export generated audio, and create Text Notes from Apple Shortcuts or Siri. • Better TTS playback: Kokoro long-form playback now keeps generated segments tied to the active session, so restarts and cleanup behave more reliably. • Safer Calendar & Reminders permissions: Actions require explicit permission, personal-data consent, and a review step before changes are applied. • Fixes and improvements: Improved Text Note selection, autosave/status feedback, keyboard behavior, and sidebar navigation.

Version 1.45.0 makes voice typing faster, files easier to add, and Knowledge Libraries more useful. • Faster voice typing: Clearer controls, quick language and model selection, Quick Paste, and live progress while you record. • Drag and drop files: Drop documents and images straight into a chat or a Knowledge Library, or paste them in. • Notes in Knowledge Libraries: Write and edit Markdown notes in the app, preview the formatting, and keep the search index current. • Re-transcribe recordings: Run a saved recording again with a different model, language, or speech enhancement, and cancel anytime. • Choose what loads at launch: Pick a model or provider to prepare on startup, or open the app with nothing loaded. • Batch actions: Select multiple chats, Voice Notes, or Text to Speech items to delete them or add them to a Knowledge Library. • More room to type on iPhone and iPad: The prompt editor expands when the on-screen keyboard is open. • Better Shortcuts and App Intents: Find and open conversations, libraries, and documents from Apple workflows. • Polish and bug fixes: Smoother navigation, clearer model status, and better attachment handling.

Version 1.44.1 makes On-Device AI easier to use globally and smoother for everyday voice work. • Multi-language support: The app now has broader language coverage. Please send suggestions or wording feedback to developer@ondevice-ai.app. • Better STT and TTS voice enhancement: Speech-to-text and text-to-speech workflows are more reliable, with improved handling around voice enhancement. • Faster chat UI: Conversation screens have been optimized for smoother scrolling and more responsive long chats. • Clearer TTS generation UI: Text-to-speech generation now gives a better visual flow while audio is being prepared. • Improved model downloads: TTS and STT model downloads are easier to follow and more stable. • Bug fixes and polish: Includes smaller fixes across voice, chat, and model workflows. • Fixed an issue where the keyboard could block the input field. • Fixed an issue that could prevent vision models from loading.

Version 1.44.0 makes On-Device AI easier to use globally and smoother for everyday voice work. • Multi-language support: The app now has broader language coverage. Please send suggestions or wording feedback to developer@ondevice-ai.app. • Better STT and TTS voice enhancement: Speech-to-text and text-to-speech workflows are more reliable, with improved handling around voice enhancement. • Faster chat UI: Conversation screens have been optimized for smoother scrolling and more responsive long chats. • Clearer TTS generation UI: Text-to-speech generation now gives a better visual flow while audio is being prepared. • Improved model downloads: TTS and STT model downloads are easier to follow and more stable. • Bug fixes and polish: Includes smaller fixes across voice, chat, and model workflows.

Version 1.43.0 brings richer voice, more control over your models, and smoother ways to manage conversations you want to keep. • Voice Cloning for More TTS Models: Several text-to-speech models now support cloned voices, so you can have AI speak in a voice that sounds like you across more engines than before. • New TTS Engines. Qwen 3 TTS, CostVoice, VibeVoice: Three new text-to-speech models are now available, giving you more tone and style options for voice playback and conversation mode. • Custom Model Save Path: You can now choose where your AI models are stored on disk. Point the app at any folder — an external drive, a shared volume, wherever makes sense for your setup. • Fork, Copy, and Export Conversations: Branching off from an earlier message is now faster, and you can copy or export full conversation threads in a few taps, useful for sharing context or keeping records. • Faster Long Conversations: Scrolling and responding in long chat threads is noticeably smoother, with reduced lag when your conversation history grows large. • UI and Experience Polish: A range of smaller refinements across navigation, layout, and interactions to make everyday use feel more consistent and responsive. Update now for more voice options, flexible model storage, and better tools for managing your conversations.

Version 1.42.0 gives voice and agent workflows more room to breathe. You get more local speech model options, a new PocketTTS voice engine, cleaner subagent editing, and the ability to fork a conversation from the middle when one answer deserves its own path. • Qwen3-ASR, Nemotron Speech, and Parakeet are now supported, giving voice notes, dictation, and conversation mode more transcription options. • PocketTTS is now available as a text-to-speech engine for natural voice playback. • You can start a new conversation from an earlier response without changing the original thread. • Editing specialist agents is easier, especially when refining instructions during larger workflows. • This release includes fixes across chat, voice, and agent flows for a steadier day-to-day experience. Update now for better speech support, easier branching, and a smoother subagent workflow.

Dictate into any Mac app, connect to more cloud providers, and let your AI handle complex tasks with less friction. Version 1.41.0 focuses on getting out of your way so you can work faster and more securely. • Enhanced IM Channel Security: Take control over who can reach the app. You can now restrict incoming IM messages to a specific list of approved users, keeping your workflow secure and uninterrupted. • Better Speaker Diarization: We've updated the logic for distinguishing between different voices, resulting in much more accurate and readable transcripts when multiple people are speaking. • Safer Credential Storage: API key credentials are now stored in your device's Keychain using enhanced security logic, ensuring your sensitive data is better protected. • Gemma 4 & Expanded Model Support: You can now connect to and run Google's newly released Gemma 4 models, alongside a handful of other recently released models, giving you even more options for offline processing and agentic workflows. • Refactored Agent Flow: Improved user experience with easier editing and management of custom agents. • Export & Import Agent Flows: You can now easily back up your custom agent setups or share them with your team by exporting and importing agent workflows. • Enhanced Chat Communication: More robust and reliable chat operations for smoother interactions. Update now to try voice typing on your Mac, share your custom agent flows, and connect to your preferred models with tighter security and fewer steps.

Dictate into any Mac app, connect to more cloud providers, and let your AI handle complex tasks with less friction. Version 1.40.0 focuses on getting out of your way so you can work faster. • Voice Typing for Mac: Press a hotkey, speak naturally, and your words appear directly in whatever app you're working in: Notes, Mail, Slack, a code editor. A small floating panel shows what you're saying in real time. AI can clean up grammar and filler words before the text is placed. No copy-paste needed. • Browser Agent: You can now see what the browser agent is doing step by step. It can pause for your input during a task, and stopping it mid-run produces a clean result instead of leaving things in a broken state. • Three New Cloud Providers: Cloudflare AI Gateway, GitHub Models, and Microsoft Foundry are now available alongside 15 existing providers. More options for teams already using these platforms. • Simpler Cloud Settings: All provider setup — browsing, credentials, model selection — now happens in one place inside App Settings. No more jumping between screens. On iPad, the layout adapts to your screen size automatically. • Smarter Hugging Face Imports: When you add a model from Hugging Face, the app reads its metadata and pre-fills settings like chat format and vision support for you. Everything stays editable, so you can adjust before saving. • Better Image Understanding: Vision models now handle photos more reliably across different devices. Switching between vision models no longer causes conflicts, and multi-agent conversations re-evaluate image handling when the active model changes. • Tool Calling: You can also set default values for tool parameters in Settings, so your AI fills in the blanks the way you prefer. Update now to try voice typing on your Mac or connect your preferred cloud provider in fewer steps.

Transform your AI into a smarter, more capable workspace with version 1.39.0. This update introduces groundbreaking tools and enhancements to make your workflows faster, more collaborative, and more efficient—all while keeping your data private. • Autonomous Subagents: Break down complex tasks with ease. Your main agent can now create specialized subagents to handle focused parts of larger workflows, making your projects more structured and scalable. • Live Browser Operation Tool: Watch your AI navigate the web in real time. Open pages, complete multi-step tasks, and take control manually when needed—all within a visible browser. • Custom MLX Model Import: Bring your own MLX models from Hugging Face with just a repository ID or URL. The app now helps detect model capabilities like vision and reasoning for seamless integration. • Streamlined iPhone Experience: Enjoy a cleaner, faster, and more intuitive interface designed for everyday mobile productivity. • Enhanced Knowledge Workflows: Pause and resume deep document analysis without losing progress. Improved API support ensures compatibility across providers. • Faster Core Performance: Experience smoother, faster operations with optimized image processing, an updated inference engine, and a wide range of bug fixes. Update now to unlock these powerful new features and elevate your AI experience. As we continue to introduce advanced capabilities, our pricing may increase in the near future. However, for a limited time, you can lock in our Lifetime Pro offer and secure all current and future updates. Version 1.39.0 is designed to help you work smarter, not harder.

Supercharge your AI experience with professional audio tools, expanded intelligence, and lightning-fast workflows. Version 1.38.0 brings massive upgrades to Voice Notes, powerful new cloud models, and deep UI refinements. • Pro Audio & Voice Notes: Never lose track of a conversation again. Pro users can now automatically identify who is speaking with Speaker Diarization (with detailed settings), and import existing audio files for high-accuracy Apple STT transcription. We've also added the ability to rename recordings, smoothed out the entire Voice Notes interface, and resolved Whisper STT loading issues. • Seamless Text-to-Speech: Listen on your terms. Instantly trigger TTS with new dedicated speaker buttons on iOS and a macOS header button. Send AI responses directly to the TTS workflow in one tap, preview voices with audio/video examples, and now hear your AI speak in your own voice with full support for Apple Personal Voice. Plus, enjoy stable, crash-free Kokoro audio generation with smart RAM management. • Expanded Intelligence: Tap into the most advanced models on the planet. We've added Mistral, xAI (Grok), and Hugging Face to our cloud providers, and introduced local support for Qwen 3.5. • Bulletproof Tool Calling: Your AI agents just got smarter. We've fixed global memory searches, refined MLX and GGUF prompting formats, prevented empty loops, and added the ability to instantly cancel running tools without freezing. • Frictionless Workflow: Get to work faster. Customize your startup screen to bypass initial model loading. Enjoy seamless iOS image sharing, improved Telegram streaming, enhanced VoiceOver accessibility, and beautifully reorganized app settings.

Supercharge your productivity with parallel AI teams and complete remote control. Version 1.37.0 transforms how you get work done by allowing multiple AI agents to tackle complex tasks simultaneously, while putting complete command of your workspace right inside your favorite messaging apps. • Parallel AI Teams: Time is your most valuable asset. Stop waiting for sequential responses. Our new Complex Task Planner allows specialized AI agents to work on multiple parts of your project at the exact same time. • Complete IM Command Center: Control your entire AI workspace from Slack or Discord without opening the app. Seamlessly switch models, swap chat flows, toggle knowledge bases, and manage tool access directly through simple chat commands, backed by robust connection stability and token-aware context truncation so chats never lag. • Advanced Tool Mastery: Customize exactly how your AI interacts with the world. Set default parameters for specific roles, enable global tool memory, and execute network requests with new cURL capabilities. We've also update web fetching to make online research cleaner and more accurate. • Expanded Cloud Horizons: Seamlessly tap into top-tier cloud models when you need maximum power. We now support AWS Bedrock, Z.ai (Zhipu GLM), Opencode Zen, Qwen Portal, and Kimi—with smarter auto-detection for vision models to make image analysis effortless. • Frictionless Experience: Focus on your work, not the setup. We've completely redesigned and streamlined the app settings, fixed tool-window freezes, and launched a comprehensive usage guide on our website to help you master every feature. Deploy AI teams that operate at the speed of thought, all orchestrated seamlessly from wherever you work. Update your On-Device AI experience today.

Turn every project into a connected AI workspace with seamless IM integration and active tools. This update brings real-time web search, advanced multi-flow chat management, and a high-speed modern speech engine directly to your device. • IM Connectivity (Mac): Connect your AI to Instant Messaging platforms (Discord, Slack, Telegram) to automate replies and manage conversations from your desktop. • Tool Calling & Web Search: Empower your agents with active tools. Integrated web search with a quick toggle brings real-time information into your chat. • Advanced Chat Flows: Manage multiple chat flows and keep different contexts organized. Now with flexible model switching for all users. • Enhanced User Experience: - Expanded Input Window: A larger workspace for defining characters and creative writing. - Optimized Interactions: Improved accessibility and control layout, particularly for spatial interactions on Vision Pro. - Cloud Provider Updates: A fresh look for cloud model settings. • Modern Speech-to-Text: Experience faster and more accurate voice transcription with the new built-in Modern Apple Speech-to-Text engine. • Responsive Feedback: Clear status indicators now keep you informed during intensive operations like high-quality speech generation, so you always know what's happening. Experience a more connected and capable AI with specialized tools, expanded connectivity, and a UI designed for creativity.

• GGUF Model Stability: Enhanced stability for GGUF models, ensuring smoother operation during offline inference • Smart Memory Management: Improved memory handling for both LLM inference and TTS generation, significantly reducing crashes on memory-constrained devices • Engine Upgrade: Updated MLX and llama.cpp engines to provided the latest performance and compatibility improvements • Adaptive TTS Engine: Automatically adjusts the Text-to-Speech engine based on available RAM during conversation mode for optimal stability • Better Keyboard Response: Optimized input handling for a snappier, more fluid typing experience during active conversations Experience a more stable and efficient local AI. Version 1.35.0 brings critical memory optimizations, engine updates, and adaptive TTS to ensure your assistant runs smoothly, even during complex tasks.

GGUF Vision Model Support • Run vision-capable GGUF models directly on your device for powerful image understanding without cloud dependency • Analyze photos, screenshots, documents, and diagrams using local vision models with complete privacy • Seamlessly switch between text-only and multimodal conversations Enhanced Camera & Vision Workflow • Updated camera mode allows capture image at free form for vision model analysis • Snap photos of real-world objects, whiteboards, or documents and get instant AI insights • More intuitive camera interface designed for quick visual queries Smarter Knowledge Libraries • Refactored embedding pipeline with session-based logic for more reliable document knowledge storage • Export and import knowledge packages to backup, share, or transfer your curated libraries between devices • Change embedding models directly from the main conversation screen for flexible document analysis Improved Audio Recording Experience • Better audio recording handling with more robust microphone management • Fixed potential microphone permission issues on iOS 26 for reliable voice capture • Smoother start/stop transitions and improved error recovery during recording sessions Whether you're analyzing images with local vision models, building portable knowledge libraries, or capturing voice notes with enhanced reliability, version 1.34.0 delivers more powerful and flexible on-device AI capabilities while keeping your data completely private.

Knowledge Libraries arrive in version 1.33.0, giving every project its own dedicated AI memory. Keep research clean, context focused, and documents exactly where your assistant expects them. Knowledge Libraries for serious projects: • Create a Library for each project or client and keep related notes, PDFs, and screenshots grouped together • Manage everything from one place, review document previews, then pull them into a conversation when you are ready • Library aware search focuses answers on the active Library for clearer, more trustable results • This release is the foundation, future updates will expand Knowledge Libraries with more powerful tools for organizing and sharing knowledge Conversation polish and engine reliability: • Scrolling in long conversations is smoother and more predictable, even with detailed exchanges • Keyboard reactions in the chat view are cleaner, reducing surprise jumps while typing and editing • Embedding model inference is more robust, with better indexing and fewer errors on large collections • Many small stability improvements make everyday work with documents and chat more dependable Whether you're managing separate research projects, client knowledge bases, or long-running personal libraries, version 1.33.0 gives you a powerful Knowledge Library system that keeps your AI organized, focused, and reliable.

Unlock powerful new ways to interact with your world in version 1.32.0. This update introduces groundbreaking vision capabilities for both local MLX models and cloud APIs, allowing you to analyze images and understand visual content like never before. We've also revolutionized web search with a new page-by-page reading mode for deeper, more accurate insights. Combined with a polished UI and a more robust MLX engine, your AI assistant is now more intelligent, perceptive, and reliable than ever. Multimodal Vision for Local & Cloud AI - MLX Vision Support: Analyze images directly on your device with select MLX models. Understand photos, diagrams, and real-world objects with complete privacy. - Take a picture of a plant and ask your local AI to identify it, all without an internet connection. - API Vision Integration: Leverage the power of advanced vision models from cloud providers. Get detailed descriptions, text extraction, and object recognition from services. - Upload a screenshot of a complex chart to get an instant summary and data analysis from a powerful cloud-based vision model. Deep Web Search with Page-by-Page Reading - Go beyond summaries with our new "Open for Web Search" feature. Your AI can now read web pages sequentially, gathering more accurate and detailed information for comprehensive research. - Ask your AI to research a complex topic, and it will read through multiple pages of a source article to give you a detailed and nuanced answer. Enhanced Performance & User Experience - Robust MLX Inference: We've hardened the MLX engine for greater stability, reducing errors and ensuring more reliable performance during long or complex tasks. - Polished UI: Enjoy a smoother, more intuitive experience with various UI updates designed to make your workflow more efficient and enjoyable. Whether you're identifying real world objects with a photo, extracting insights from complex charts, or conducting comprehensive research across the web, version 1.32.0 equips your AI assistant with the power of sight and deeper analytical intelligence. Experience the most perceptive and reliable On-Device AI yet.

Enhanced RAG Intelligence & Precision - Advanced embedding pipeline with improved semantic understanding and chunk scoring for more accurate document retrieval - Smart token budget calculations based on actual embedding model context sizes and chunk dimensions - Hybrid search combining vector similarity with text matching for comprehensive knowledge base queries - Research Workflows: Get more relevant citations and fewer off-topic results when analyzing multi-document knowledge bases - Academic Projects: Better context extraction from research papers with improved semantic grouping and relevance scoring Optimized Web Context Analysis - Cleaner webpage extraction with enhanced noise reduction and content structure preservation - Improved handling of dynamic content, JavaScript-rendered pages, and multi-step loading sequences - Better metadata capture including titles, authorship, and document structure for richer context - Web Research: Generate more focused summaries with reduced duplicate content and boilerplate text elimination - News Analysis: More accurate extraction of article content Performance & Stability Improvements - Smoother UI interactions with reduced memory usage and optimized state management across conversation screens - Resolved critical inference issues with GGUF model execution. - Enhanced GGUF model reliability with better error recovery and stable context handling - Improved embedding model loading with proper resource cleanup and memory management - Extended Sessions: More stable performance during long research projects with continuous document analysis and web browsing Transform your research and analysis workflows with smarter document understanding, cleaner web content extraction, and rock-solid performance. Whether conducting academic research with complex knowledge bases, analyzing web content for insights, or managing extended AI sessions, version 1.31.0 delivers the most intelligent and reliable On-Device AI experience yet.

Streamlined Model Import & Discovery - Import GGUF models directly from Hugging Face repositories with a simple URL or model identifier - Enhanced local model import with improved file handling, better error detection, and validation - Discover and download cutting-edge models from the community without leaving the app - Research Workflows: Quickly experiment with newly released models from Hugging Face for specialized tasks - Custom Solutions: Import organization-specific fine-tuned models directly from private or public repositories Enhanced Productivity Shortcuts - Expanded Mac hotkey system with customizable shortcuts for quick access to different app sections - iOS Quick Actions for launching directly into chat, voice recording, or document analysis from the home screen - Jump to specific features instantly without navigating through menus - Professional Workflows: Configure hotkeys for frequently used features like new conversation, voice recording, or model switching - Mobile Efficiency: Long-press the app icon on iOS to immediately start voice recording or open a specific tool Performance Optimizations & UI Refinements - Faster model loading times with optimized initialization sequences and improved memory allocation - Refined user interface with smoother animations, better visual hierarchy, and enhanced responsiveness - Reduced battery consumption during extended AI sessions Model Management Improvements - Better model validation during import with clear error messages and recovery suggestions - Enhanced compatibility detection to ensure imported models work optimally on your device - Improved model organization and categorization in the model picker - Quality Control: Automatic verification ensures imported models are compatible before adding to your library - Smart Recommendations: Get device-specific suggestions when importing models based on your hardware capabilities Experience seamless access to the world's largest AI model repository with direct Hugging Face integration, combined with lightning-fast shortcuts that put powerful AI tools at your fingertips. Version 1.30.0 makes it easier than ever to discover, import, and use cutting-edge AI models while maintaining the privacy and performance you expect from on-device processing.

Cloud Provider Integration - Connect to any compatible third party or self hosted API endpoint with your own key for access to powerful cloud models. - Seamlessly switch between on-device and cloud models within the same conversation for optimal performance and privacy balance - Business Use: Access latest Cloud Model for complex analysis while keeping sensitive data local on your device - Flexible Workflows: Use cloud models for heavy reasoning tasks and local models for quick responses or offline scenarios Enhanced Session Management - Remove individual attached documents from conversations with a simple tap on the close button - Clean up document references without affecting previous conversation responses - Research Projects: Dynamically manage your knowledge base by removing outdated PDFs or irrelevant images mid-session - Document Review: Add multiple files for comparison, then remove unnecessary ones to focus AI analysis on relevant materials Improved Inference Stability - Resolved memory management issues when switching between local and cloud models for smoother transitions - Enhanced error handling and recovery during model loading and inference processes - Multi-Model Workflows: Switch between local models and cloud APIs without performance degradation - Extended Sessions: More stable operation during long conversations with frequent model switching UI Refinements & Bug Fixes - Fixed display issues in model picker and settings interfaces for better user experience - Improved visual feedback and state indicators throughout the application - Streamlined Settings: Cleaner interface for managing both local and cloud model configurations - Better Discovery: Enhanced UI makes it easier to explore and configure cloud provider options Experience the perfect balance between privacy and power. Connect to cutting-edge cloud models when you need maximum capability, while keeping your sensitive data secure with local processing. Version 1.29.0 delivers the most flexible and stable AI experience yet, with seamless cloud integration and refined session management.